The three main methods to implement molecular polarization (point dipoles, fluctuating charges, and shell model) are tested against high level ab initio calculations for a molecule (water, carbon tetrachloride) close to a point charge (at the distance of a lithium or magnesium ion). The goal is to check whether an approximation (linear polarization) strictly valid at large intermolecular distances is sufficiently accurate for liquid state molecular dynamics simulations, where strong polarization effects are to be expected at short separations. The monitored observable is the molecular dipole moment as a function of the charge-molecule distance for selected molecular orientations. Analytic formulas are derived for the components of the molecular polarization tensor, facilitating the optimization of the performance for each polarization method as a function of its underlying parameters. Overall, the methods studied provide a remarkably good representation of the induced dipole, with no divergences appearing even at the shortest distances. For water close to a monovalent point charge the point dipole model, implemented with one or three dipoles, accurately reproduces the water dipole moment at all distances. Deficiencies appear as the molecular polarizability and/or charge increase: basically, the ab initio induced moments grow faster at intermediate distances than the linear increase characteristic of the phenomenological polarization methods, suggesting that nonlinear effects (hyperpolarizability) cannot be neglected in these cases. Regarding the capabilities of each method, the point dipole method is the one that performs best overall, with the shell model achieving acceptable results in most instances. The fluctuating charge method shows some noticeable limitations for implementations of comparable complexity (in terms of the number of sites required). (C) 2004 American Institute of Physics.